Skip to content

Latest commit

 

History

History
44 lines (31 loc) · 1.51 KB

File metadata and controls

44 lines (31 loc) · 1.51 KB

Executive Summary

One-Sentence Thesis

Modern coding agents do not share one standard implementation, but they are increasingly converging on a reusable runtime pattern.

The Short Version

The strongest current candidate for that pattern is a guarded tool-result loop:

  1. the agent works through explicit tool calls,
  2. tool results are fed back into the next turn,
  3. failures remain recoverable and in-band,
  4. completion is explicit and must be validated,
  5. runtime state is inspectable outside the transcript,
  6. context compaction and resumability are runtime concerns,
  7. hooks and policy layers sit around the loop rather than replacing it.

What Looks Closest To Standardization

  • tool-result loop,
  • explicit completion attempts,
  • guarded completion,
  • recoverable tool failures,
  • command-output streaming,
  • minimal runtime state exposure.

What Still Looks Unstable

  • journaling format,
  • exact compaction protocol,
  • stop-hook standardization,
  • subagent orchestration details,
  • durable resume and checkpoint interfaces.

Current Opinion

If this repo had to be opinionated today, it would say that the right abstraction layer is neither the raw SDK layer nor a full workflow engine. It is a narrower runtime layer that defines task lifecycle, turn execution, completion gating, failure reinjection, state, compaction, and bounded delegation.

Read Next

  1. patterns/README.md
  2. standardization-view.md
  3. runtime-contract.md